Speak machine

A

Actuator

The motor-and-transmission unit that moves a joint — the robot equivalent of a muscle. Actuator count, torque and quietness set what a machine can physically do, and they dominate its bill of materials, which is why prices fall only as fast as actuators get cheap.

Android / gynoid

An android is a robot built to look human, not just human-shaped; a gynoid is the female-presenting kind. Most of today's companion market is, strictly speaking, gynoids — the industry mostly says 'humanoid' for everything.

Animatronics

Pre-scripted mechanical performance — the theme-park and film tradition of lifelike motion without intelligence. Many convincing companion demos are closer to animatronics than autonomy: gorgeous playback, no understanding. The paper trail exists to tell those apart.

B

Bipedal locomotion

Walking on two legs — still one of the hardest problems in robotics, demanding constant balance recovery under shifting loads. It's why the most expressive companion faces tend to ride wheeled bases, while the best walkers have industrial faces.

Bust (form factor)

A head-and-shoulders companion without a body — the cheapest way to buy a lifelike face and conversation. Several makers sell the same head as a bust or atop a full body at very different prices.

C

Cloud dependence

How much of a robot's mind lives on the maker's servers. A cloud-dependent companion can lose conversation, memory or features if the subscription lapses or the company folds — which is why we flag it on every profile.

D

Degrees of freedom (DoF)

The number of independent ways a machine can move — each powered joint axis counts as one. More DoF means more expressive motion, but spec-sheet DoF says nothing about how naturally those joints are coordinated.

E

Embodied AI

AI that perceives and acts through a physical body rather than a chat window. The bet behind the whole field: that presence — a face, a voice in the room, something that turns to look at you — changes what an AI relationship is.

End effector

The business end of a robot arm — in humanoids, the hand. Hands are notoriously the hardest and most expensive subsystem; more than one shipping delay in this directory traces back to a hand redesign.

F

Facial actuation

The motors under a robot's face. Count them: a companion with 17+ facial motors can attempt micro-expression; one with three can smile, roughly. Facial actuation is the main input to the face dimension of our realism index.

H

Human-in-the-loop

A system where a person supervises or completes what the machine can't — approving actions, taking over on failure. In home humanoids it's the honest middle ground between demo and autonomy, and some makers openly ship with it. See: teleoperation

Humanoid

Human-shaped, not necessarily human-looking. Warehouse humanoids have camera-head faces and no skin; companions trade payload for silicone and eye contact. This site tracks both, because the technology flows from one to the other.

L

Large language model (LLM)

The class of AI model behind modern robot conversation. Which model a companion runs — and whether it runs on-device or in the cloud — shapes its memory, latency, personality and privacy more than any mechanical spec.

Lip sync

Matching mouth movement to speech in real time. Millisecond-level mismatch is one of the fastest ways a lifelike face falls into the uncanny valley — it's scored inside the voice dimension of our index.

P

Paper trail

Our term for the dated, sourced record of what a maker promised versus what actually happened — prices as announced, ship windows as stated, deliveries as verified. Append-only: we add to it, we don't rewrite it. See the site-wide ledger

R

Realism Index

Our 0–100 editorial score for how close a machine comes to lifelike human presence — six weighted dimensions (face 25%, body & skin 20%, motion 20%, voice 10%, mind 15%, reality 10%), scored from public evidence only. Read the methodology

S

Silicone (medical-grade)

The premium skin material of the lifelike end of the market — durable, paintable in layers, and temperature-neutral to the touch unless actively heated. The realistic look comes as much from pigmentation labor as from the material itself.

Sim-to-real

Training a robot's skills in physics simulation, then transferring them to hardware. It's how modern humanoids learned to walk years faster than their predecessors — and why capabilities can improve by software update.

T

Teleoperation

A human remotely piloting the robot. Essential for data collection and honest as a disclosed stopgap — but undisclosed teleop is the oldest trick in the demo playbook, and more than one 'autonomous' showpiece in our paper trails turned out to have a person inside.

TPE (thermoplastic elastomer)

The cheaper, softer alternative to silicone skin. It feels more yielding but degrades faster, stains and is harder to repair — the usual material trade at the budget end of lifelike bodies.

U

Uncanny valley

The dip in comfort when something looks almost — but not quite — human, first described by roboticist Masahiro Mori in 1970. Stillness flatters; motion betrays. It's the single force that shapes this entire product category. The full guide

V

Vision-language-action model (VLA)

A single AI model that maps what a robot sees and is told into physical action — the architecture behind the current generation of general-purpose humanoids. It's why makers now talk about robots the way they talk about chatbots: trained, not programmed.

Voice synthesis (TTS)

Text-to-speech — the generated voice a companion speaks with. Modern TTS is near-human in isolation; the remaining tells are conversational latency and how the voice couples with the face.

W

World model

An AI system's internal prediction of how the physical world behaves — what happens next if it moves, pushes, releases. Better world models are the research bet for getting robots off teleoperation and into genuine autonomy.